Google Warns Hackers Are Turning AI Agents Into Attack Tools
Attackers are using AI tools to automate parts of hacking, like scanning networks and stealing passwords, with less human help. Google warns this could speed up attacks and make them harder to stop.
- Report priority
- High
- Involves
What is known
- Attackers now give AI tools specific jobs, like searching for weak spots, fixing problems in hacking code, or stealing passwords, so the AI can do many steps of an attack without much human input.
- Before, attackers would ask AI for one thing at a time, like writing code or crafting fake emails, then use that output themselves.
- Now, AI can handle multiple parts of an attack, like checking a company's defenses, troubleshooting issues, and moving through the attack plan on its own.
- This makes attacks faster and harder to stop because the AI can keep working even if a human isn't constantly guiding it.
What to do
If your company or organization uses AI tools to automate security tasks, like scanning networks, troubleshooting, or managing workflows, check if your security team is using AI agents to handle parts of hacking workflows, this could make attacks faster and harder to detect. If you're not sure, ask your IT or security team if they're using AI tools in their processes.
If your organization uses AI tools for security or automation, review how they're being used and ensure they're not being misused for malicious purposes. Work with your IT or security team to monitor AI activity and set up alerts for unusual behavior. For now, there's no direct fix, but staying aware of AI risks can help protect your systems.
Reported details
Attackers use AI to scan a company's network for weak points, then automatically try to steal passwords or spread malware without needing constant human direction.
Google's Threat Intelligence Group documented attackers deploying AI agents to automate multiple stages of cyberattacks with minimal human oversight. Unlike traditional AI-assisted tasks, such as writing malware or crafting phishing emails, these agents now coordinate operations like credential harvesting, infrastructure scanning, and attack workflow management. In one observed case during Q2 2026, threat actors completed a mass credential harvesting campaign in under six hours using AI-driven automation.
While fully autonomous attacks remain rare, the trend suggests attackers are shifting toward agentic systems that reduce reliance on manual intervention. Beyond using AI as a tool, attackers are also weaponizing compromised cloud infrastructure. Legitimate cloud resources are being repurposed to route malicious traffic, obfuscate attacks within normal cloud traffic, and avoid building their own infrastructure.
This approach expands the attack surface beyond traditional targets, leveraging cloud services as both a staging ground and an enabler of broader malicious operations.